(This agent models everything as a game tree
 (there is the space of all possible commands
  (there is the space of commands the system knows how to construct)
  (there is the space of commands the system understands the effects of)
  (the system knows what a string is (cyc))
  (the system has classical AI planning capabilities (pddl))
  (the system learns the effects of actions by making
   observations, performing actions, and observing what changes) 
  (it can also be instructed as to what happened)
  )
 (so I guess I have to mount various systems here)
 (the system should be able to learn based on experience and observation, such as, memoizing the results of function calls)
 (keeps a game tree as to what happens if you do this or that in this particular situation.)
 
 (huge explosion of information, so how do we comfortably and compactly render this?)
 
 (look into exploration algorithms, type theory, etc.)

 (tutorials, system reads through tutorials)

 (sensors and effectors)

 (makes observations about program output, patterns, etc)

 )

(eventually system self-hosts)

(need to come up with an efficient way to make observations about
 text, maybe using that string matching algorithm by Knuth)

(Maybe start by looking at platonic reality and mapping that out,
 using algorithms for generating axioms and axiom schemas, proof
 systems, and seeing what things we can prove about those
 different systems from stronger systems.  Develop a sense about
 proving things in these languages.)

(prolog-agent should have the option of controlling a vagrant or
 a regular computer)
(figure out the vagrant commands for returning to a checkpoint)